Indiegogo: Checkpoint #2
Data Analytics: R @ Hult IBS
Raafi Jahangir, Rodrigo Landeros
Set-up
# Install "fastDummies" package to quickly create dummy variables (UN-COMMENT THE CODE TO DOWNLOAD THEN COMMENT AGAIN ONCE DOWNLOADED SO IT DOESN'T RUN AGAIN)
#install.packages('fastDummies')
# package that share common data representations and 'API' design
library(tidyverse)
# package to create visualizations
library(ggplot2)
# package called "broom" that converts statistical objects into Tiny Tibbles
library(broom)
# package called "fastDummies" is used to create the dummy variables all at once, saving time. We just select the Dummy variables that we need.
library('fastDummies')Abstract
As part of the Data Analytics: R class, teams were tasked with identifying a business need and creating an analysis to try to solve the business need.
The data that was chosen by our team is based on the success rate on projects that are uploaded to Indiegogo, an online platform that serves as a crowd funding for projects, products and services.
Topic
Success rate of Indiegogo campaigns categories in the last 10 years.
Introduction
When it comes to crowd funding websites, we have always heard about Kickstarter and Indiegogo. Grasshopper Resources shares a great 3-way comparison between Kickstarter, GoFundMe and Indiegogo. For this case, we will be taking only Kickstarter and Indiegogo into consideration. The article shares a snapshot the pros and cons for both platforms, see below:
Funded Today is an article that sides Kickstarter since the very beginning but mentioned about some of the advantages of Kickstarter including exceptional customer service and allow pre-prototype launch. These are two great attributes to have for students since we are still a beginner and just need to put our foot in the door. We also wanted to choose the underdog in this case and learn more about its campaigns, categories, amount funded, and percentage funded.
Objectives
We want to find out if Indiegogo campaigns in a specific category have different success rate compared to other categories. As business analytics students and aspiring entrepreneurs, we want learn more about what categories on the platform have the highest success rates, most amount of money raised, and the average duration of campaign length. These things will allow us to gear our future projects into this direction to guarantee whether or not our projects will be successful on this platform. This analysis will cover all of these elements to allow us and our fellow classmates to make an informed decision on whether or not to use Indiegogo to crowd fund our future projects.
Hypothesis:
H0 = Indiegogo campaigns have the same success rate compared to other categories on the platform
H1 = Indiegogo campaigns have different success rates compared to other categories on the platform
Plan
We will check the success rate of the campaigns over the last few years to check if the chances of fully fund a project have been the same or have increased/decreased.
We will be finding out the amount of successful projects per category and how they rank with other categories
We will be creating box plot, histograms, scatter plots, and bar graphs to visualize our analysis.
We will create an analysis based on the top five categories and will try to explain why they are the leading categories compared to the ones that aren’t.
Data
Background
This data set was obtained from an online scraper that scraped Indiegogo’s campaigns for the last few years. We will be only using this year’s data that is a snapshot of the last 10 years as there are several campaigns
open_datefrom 2010.The data set source is hyper-linked. We individually downloaded the
csvfiles from each months in 2020.
Data Cleaning
There were multiple csv files per month and we obtained the last 11 months of snapshots for the past decade worth of data. This made data loading difficult. We were able to collectively import all data sets and merge it into one data set.
Just to double-check if our file is running correctly, we can run a summary and check its length, class, and mode. For the numeric values, we can see its basic descriptive statistics.
# data summary
summary(master_file)## bullet_point category category_url clickthrough_url
## Length:708648 Length:708648 Length:708648 Length:708648
## Class :character Class :character Class :character Class :character
## Mode :character Mode :character Mode :character Mode :character
##
##
##
## close_date currency funds_raised_amount funds_raised_percent
## Length:708648 Length:708648 Min. : 0 Min. : 0.0
## Class :character Class :character 1st Qu.: 255 1st Qu.: 0.0
## Mode :character Mode :character Median : 1200 Median : 0.2
## Mean : 55537 Mean : 27.9
## 3rd Qu.: 5680 3rd Qu.: 1.0
## Max. :51486048 Max. :1641272.0
## image_url is_indemand is_pre_launch offered_by
## Length:708648 Mode :logical Mode :logical Length:708648
## Class :character FALSE:656844 FALSE:705434 Class :character
## Mode :character TRUE :51804 TRUE :3214 Mode :character
##
##
##
## open_date perk_goal_percentage perks_claimed price_offered
## Length:708648 Length:708648 Length:708648 Length:708648
## Class :character Class :character Class :character Class :character
## Mode :character Mode :character Mode :character Mode :character
##
##
##
## price_retail product_stage project_id project_type
## Length:708648 Length:708648 Min. : 3036 Length:708648
## Class :character Class :character 1st Qu.:1341182 Class :character
## Mode :character Mode :character Median :2138419 Mode :character
## Mean :1872337
## 3rd Qu.:2476348
## Max. :2642275
## source_url tagline tags title
## Length:708648 Length:708648 Length:708648 Length:708648
## Class :character Class :character Class :character Class :character
## Mode :character Mode :character Mode :character Mode :character
##
##
## Number of columns and rows:
# number of observations (rows)
nrow(master_file)## [1] 708648# number of variables (columns)
ncol(master_file)## [1] 24First five rows:
# first five rows
master_file %>% slice_head(n=5)Removing duplicates
This is currently a still dirty data set meaning it needs to be cleaned. We will be cleaning in several ways. The first one is to remove duplicate observations.
# remove duplicates based on project_id
indiegogo <- master_file %>% distinct(project_id, .keep_all = TRUE)We just removed all duplicates by project_id so we now have an individual project_id per campaign in the clean data set.
# number of rows (observations)
nrow(indiegogo)## [1] 42902There were 708648 observations in the dirty master_file, and now there are 42902 observations the new data set indiegogo. It is a huge decrease, but exactly how much?
# row change
nrow(master_file)/nrow(indiegogo)## [1] 16.51783The new data set was cut by more than 16 times! That means that, on average, there were 16 or 17 duplicates per campaign.
Dropping columns
The next cleaning step is to remove columns that are not needed and won’t be taken into consideration for the analysis.
# drop unneeded columns
indiegogo <- indiegogo[, !names(indiegogo) %in% c("bullet_point","category_url",
"source_url","image_url",
"offered_by","perk_goal_percentage",
"perks_claimed","price_offered",
"price_retail","project_type",
"tagline", "is_indemand",
"is_pre_launch", "product_stage", "tags")]We dropped 15 variables (columns) and see that the new data set now only has 9 variables (columns). This definitely cleaned up the data set to our liking, but we’re not done yet.
Filling NAs
We will also be filling values with NA if they are null.
# replace all values containing "null" to actual null value
is.na(indiegogo) <- indiegogo == "null"Now all values that were entered as “null” has been nullified.
Removing unwanted NA values
We looked closer into the data set and found out that some of the open_date, close_date and title were null. In these cases, we plan to remove them from the data maintain a clean data set
# removing rows where there is a null value in the selected columns
indiegogo <- indiegogo[complete.cases(indiegogo[, c("open_date",
"title",
"close_date")]), ]We see that a little over 400 observations were removed and we are left with 42492 observations (rows) and 11 variables (columns)
Removing dates that does not match / open projects
Something we noticed was that some of the dates open_date and close_date had a different format. In these cases, we plan to remove them from the data maintain a clean data set so we can make calculations in the future. Another aspect we wanted to check is that open projects might skew the information because they are increasing its funding each day that passes, so we took the decision to remove them to get and “static” view.
#filtering out invalid dates in the close and open date
indiegogo <- subset(indiegogo, nchar(as.character(close_date)) == 25)
indiegogo <- subset(indiegogo, nchar(as.character(open_date)) == 25)After this our ‘final’ dataset contains 40,039 rows and is ready for us to start working on it.
Mutuating and creating new columns
Now we will create some extra columns that will aid us with the project and that we will use in future stages of the project.
# We will add 9 columns that we are going to use later on.
indiegogo <-
indiegogo %>%
mutate(open_date = strtrim(open_date, 10),
close_date = strtrim(close_date, 10),
year_close = substr(close_date, 1, 4),
year_open = substr(open_date, 1, 4),
date_diff = as.double(as.Date(as.character(indiegogo$close_date), format="%Y-%m-%d")-
as.Date(as.character(indiegogo$open_date), format="%Y-%m-%d")),
funds_raised_100_percent = round(funds_raised_percent * 100, digits = 2),
goal_achieved = ifelse(funds_raised_100_percent >= 100,1,0),
log_percent = log(funds_raised_100_percent),
funds_dollars =
ifelse(currency == "AUD",funds_raised_amount/1.31,
ifelse(currency == "CAD",funds_raised_amount/1.28,
ifelse(currency == "CHF",funds_raised_amount/0.88,
ifelse(currency == "DKK",funds_raised_amount/6.07,
ifelse(currency == "EUR",funds_raised_amount/0.82,
ifelse(currency == "GBP",funds_raised_amount/0.74,
ifelse(currency == "HKD",funds_raised_amount/7.75,
ifelse(currency == "NOK",funds_raised_amount/8.59,
ifelse(currency == "SEK",funds_raised_amount/8.24,
ifelse(currency == "SGD",funds_raised_amount/1.33,
funds_raised_amount)))))))))),
log_amount = log(funds_dollars))Key Variables
We just cleaned the data set and are left with 11 variables (column) which are:
colnames(indiegogo)## [1] "category" "clickthrough_url"
## [3] "close_date" "currency"
## [5] "funds_raised_amount" "funds_raised_percent"
## [7] "open_date" "project_id"
## [9] "title" "year_close"
## [11] "year_open" "date_diff"
## [13] "funds_raised_100_percent" "goal_achieved"
## [15] "log_percent" "funds_dollars"
## [17] "log_amount"Each of this has their own significance in the analysis.
category -> Campaign category (our main independent variable), it refers to the segment that the project is aimed at.
clickthrough_url -> Campaign URL on Indiegogo’s website.
close_date -> Campaign end date (used to find out how long the campaign was going on for).
currency -> Currency in which the campaign was funded (we need it to convert to dollars).
funds_raised_amount -> The amount of funds the campaign raised over its duration on the website.
funds_raised_percent -> The campaign’s percentage of completion; if it is 1 or higher, the campaign has been successful because it has raised as much as they initially asked for (our main dependent variable) .
open_date -> Campaign start date (used to compare with close_date to find out the duration of the campaign).
project_id -> Unique campaign ID (primary key).
title -> Campaign title as published on the Indiegogo website.
year_close -> An extraction of the year from close_date.
year_open -> An extraction of the year from open_date.
date_diff -> A calculation to get the campaign duration, from open_date to close_date.
funds_raised_100_percent -> We formatted the column funds_raised_percent by multiplying it by 100, that way we can see the percentages as whole numbers instead of decimals.
goal_achieved -> A flag of 0 or 1, if the project achieved 100% or more in the column funds_raised_100_percent, it will receive a 1 else, it will have a 0.
log_percent -> A mutation to the funds_raised_100_percent to log the number.
funds_dollars -> A rate conversion from currency to dollars, that way the data is not thrown off by a stronger or weaker currency.
log_amount -> A mutation to the funds_dollars to log the number.
Exploratory data analysis
We aim for 7 plots/ that will allow us to have an initial insight of the data.
Figure 1 - Histogram 1
This is a histogram shows the funds_raised_percent of the projects, as we can see, it’s not very helpful, this is why on the second graph we use the log of the same column.
indiegogo %>%
ggplot(mapping = aes(funds_raised_percent)) +
geom_histogram()## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.
Figure 2 - Histogram 1.2
By transforming the data, we can see the distribution of the funds_raised_percent in a much clearer way.
indiegogo %>%
filter(funds_raised_percent > 0) %>%
ggplot(mapping = aes(log(funds_raised_percent))) +
geom_histogram()## `stat_bin()` using `bins = 30`. Pick better value with `binwidth`.
## Figure 3 - Bar Chart
With this graph we aim to check how may projects were launched by year.
# open projects Bar Chart
indiegogo %>%
group_by(year_open) %>%
summarise(count = length(year_open)) %>%
ggplot(aes(x = year_open, y = count)) +
geom_col() +
labs(y="Number of Projects", x="Year",
title = "Number of Projects Launched by Year",
fill = "")+
theme(legend.position="top") +
theme(plot.title = element_text(hjust = 0.5))## `summarise()` ungrouping output (override with `.groups` argument)
We can observe an increasing trend as the years go by, there is an abnormality with the year 2015 but the overall trend follows an increase in the project submitted to the page.
Figure 4 - Line graph
This is a line graph of funds_raised_100_percent. This was also part of our initial visualization to see the spread of our data.
# log(funds_raised_100_percent) line graph
indiegogo %>%
filter(funds_raised_100_percent > 0) %>%
group_by(year_open) %>%
summarise(mean_success = mean(log_percent)) %>%
ggplot(aes(x = year_open, y = mean_success, group = 1)) +
geom_line() +
labs(y="Success of Projects (in %)", x="Year",
title = "Success of Projects by Year",
fill = "")+
theme(legend.position="top") +
theme(plot.title = element_text(hjust = 0.5))## `summarise()` ungrouping output (override with `.groups` argument)
Why?
These initial visualizations gave us an idea of how our data looks like. We can conclude that we have many outliers that are messing up the rest of the data. We will be looking into this deeper through the series of visualizations below.
Figure 5 - Box plot
We decided to filter out funds_raised_percent to > 0 in order to make the log_percent work. This is the visualization outcome, in order of highest median.
# funds_raised_percent vs category box plot v1
indiegogo %>%
filter(funds_raised_100_percent > 0) %>%
ggplot(aes(x=log_percent, y=reorder(category, log_percent, FUN = median, .desc =F))) +
geom_boxplot() +
labs(y="Category", x="Log Funds Raised (%)",
subtitle="Funds Raised vs Category Box Plot")
We can observe that Audio, Tabletop Games, Camera Gear, Home, and Travel and Outdoors are out top performing categories, their median is higher than the other categories, so we start to see that the category has an effect on the success of the project. We can also observe that for out top categories, the first quartile is very long, this could mean that is difficult fore the projects to take off, and the third quartile is relatively short against the other ones, this supports the idea that once a project starts to get funding, is easier to continue raising funds.
Figure 6 - Mean Plotting with filtered infomration
Creating a new data frame with only the mean values of the top five most successful categories:
# mean of successful funds percent raised per category
mean_indiegogo <- indiegogo %>%
group_by(category) %>%
summarise(mean_percent = (mean(goal_achieved)*100)) %>%
slice_max(mean_percent, n=5)## `summarise()` ungrouping output (override with `.groups` argument)#mean_indiegogo
# reorder categories in a descending fashion
mean_indiegogo$category <- with(mean_indiegogo, reorder(category, -mean_percent))
mean_indiegogo# Top 5 - Mean & Median Success Rate (%) per Category bar graph
mean_indiegogo %>%
ggplot(aes(x=category, y=mean_percent, fill = "blue"))+#, color = variable)) +
geom_bar(stat="identity", position ="identity", alpha = 0.5) +
scale_color_manual(values=c("blue"), guide = FALSE) +
scale_fill_manual(labels = c("Mean"), values=c("blue")) +
geom_hline(yintercept = 50, color = "black") +
labs(x="Category", y="Success Rate (%)",
title = "Top 5 - Mean Success Rate (%) per Category",
fill = "")+
theme(legend.position="top") +
theme(plot.title = element_text(hjust = 0.5))
In this graph, we can observe that although projects seem to be very successful meaning that the success rate averages around 50%. We can observe that launching a project on any of these top five categories has a high probability of being successful, but further analysis is required to verify this claims.
Figure 7 - Pareto Chart
Our final figure is a Pareto Chart, which is essentially a bar graph and a line graph (with cumulative values) in one, this shows the volume of the projects depending on the duration of the project and what percent it is from all the projects. This makes it easier to understand were the bulk of the information is located.
pareto_indiegogo <-
indiegogo %>%
group_by(date_diff) %>%
summarise(count = length(date_diff), .groups = 'drop') %>%
arrange(desc(count)) %>%
mutate(., cumulative_ = cumsum(count),
cumulative = (cumulative_ / tail(cumulative_, n=1))*100)
scaleRight = tail(pareto_indiegogo$cumulative, n=1)/head(pareto_indiegogo$count, n=1)
pareto_indiegogo %>%
slice_head(n=15) %>%
ggplot( aes(x=reorder(date_diff,-count))) +
geom_bar(aes(y=count), fill='deepskyblue4', stat="identity") +
geom_path(aes(y=cumulative/scaleRight, group=1),colour="red", size=0.9) +
geom_point(aes(y=cumulative/scaleRight, group=1),colour="red") +
scale_y_continuous(sec.axis = sec_axis(~.*scaleRight, name = "Cumulative (%)")) +
theme(axis.text.x = element_text(angle=90, vjust=0.6)) +
labs(title="Pareto Chart", subtitle="Projects by Duration", x="Background Source", y="Num of Projects by Duration")
After playing with the data and doing some research, we came upon that the standard predefined project length by Indiegogo is 60 days, this means that if the person responsible of submitting the project fails to set a specific length, Indiegogo will set the duration to 60 days. This translates to 70% of the campaigns are located in just 5 chosen project lengths (60, 30, 45, 1, 40 days).
Now, what about outliers? Indiegogo can allow the owner of the project to extend the length of the campaign, but how much? Well, we have projects that surpass the 700 day mark, of course they are outliers but as we can see, increasing the length of the campaign does not mean that the project will reach it desired goal. In the other graphs, we saw data that was way off, this could mean that a specific project, had a lot of attention and raised a lot of money, possibly skewing the data.
Analysis plan
Based on the graphs that we have presented, we can clearly see the the success of the project is dependent on the category that the project is targeted to. We can also notice how the top performing categories have a closer difference between them than the lower performing categories.
In a later stage, we want to have a deep dive into how the outcome if the project is affected by the category selected vs the other ones. In order to achieve this, we will be using dummy variables, by doing so, we can isolate data and view it in a more comprehensive way.
In order for us to prove or reject the null hypothesis, we will run different correlations and to determine if there is a significance in the data or not.
Issues
We are unsure that we have all the projects that have been launched in the platform since the beginning of our data, this would mean that we might be missing some helpful insights from those projects.
We don’t know if the scraping was correctly done, if the data did not meet our criteria (date formats and NA values), the rows were dropped, so we might have a much more smaller database from the one we started. We would have wanted to have an updated date column so we could track the progress of the different projects, this could have enabled us to offer a deeper dive on the performance of the projects.
We are unsure about the outcome of the project but we will overcome it by working on the project.
Modifying the Initial Hypothesys
After the initial analysis, we decide we wanted to have a deeper dive into the duration of the projects, the top 5 categories and the amount that needs to be asked to have a successful campaign.
H0 = The duration of the Indiegogo campaigns is not related to the success rate. H0 = There is no difference on the amount needed on the Indiegogo campaigns to make it successful. H0 = The top five performing Indiegogo categories have the same opportunities on being successful.
H1 = The duration of the Indiegogo campaigns is related to the success rate. H1 = There is a difference on the amount needed on the Indiegogo campaigns to make it successful. H1 = The top five performing Indiegogo categories does not have the same opportunities on being successful.
Checkpoint #2
Overlapping the Top 5 Categories with the log_percent
We want to compare how the top 5 categories perform regarding the distribution of the log_percent.
indiegogo %>%
filter(category %in% c("Audio", "Tabletop Games", "Home", "Camera Gear", "Travel & Outdoors")) %>%
filter(funds_raised_100_percent > 0) %>%
ggplot(aes(x = log_percent)) +
geom_density(aes(colour = category)) +
labs(x="Success Percent", y="Density",
title = "Success Rate by Category Distribution",
fill = "")+
theme(legend.position="top") +
theme(plot.title = element_text(hjust = 0.5))
We can observe that the length of the bell is very similar between the categories as well as the shape, the density is what is changing and is understandable since we don’t have the same amount of projects for each category.
Overlapping the Top 5 Categories with the log_amount
We want to see not how the log amount react.
indiegogo %>%
filter(category %in% c("Audio", "Tabletop Games", "Home", "Camera Gear", "Travel & Outdoors")) %>%
filter(funds_raised_amount > 0) %>%
ggplot(aes(x = log_amount)) +
geom_density(aes(colour = category)) +
labs(x="Success in Dollars", y="Density",
title = "Success in Dollars by Category Distribution",
fill = "")+
theme(legend.position="top") +
theme(plot.title = element_text(hjust = 0.5))
We can see that all of our top performing categories are bimodal, in fact, most of the other “not top performing” categories are unimodal. We are going to do some regressions to try to get to the buttom of this behavior.
Regression
Success Rate
Since our independent variable is a categorical variable, we will have to create dummy variables for each of them to run a successful regression model. We will be using the fastDummies libary to help us quickly do that automatically. The dummy dataframe has been named indiegogo_dummy and will be used to do regressions.
indiegogo_dummy <- dummy_cols(indiegogo, select_columns = 'category')The first regression we are running is the log_percent versus all of our `category to see if they are significant values.
# Project_success vs all dummy categories
indiegogo_dummy %>%
# The reason why we have to filter for funds_raised_100_percent > 0 is because if funds_raised_100_percent == 0, the log_percent will be NA and will produce an error
filter(funds_raised_100_percent > 0) %>%
lm(formula = log_percent ~
category_Art +
category_Audio +
`category_Camera Gear`+
category_Comics +
category_Culture +
`category_Dance & Theater`+
category_Education +
`category_Energy & Green Tech` +
category_Environment +
`category_Fashion & Wearables` +
category_Film +
`category_Food & Beverages` +
`category_Health & Fitness`+
category_Home +
`category_Human Rights` +
`category_Local Businesses` +
category_Music +
`category_Phones & Accessories`+
category_Photography +
`category_Podcasts, Blogs & Vlogs` +
category_Productivity +
`category_Tabletop Games` +
category_Transportation +
`category_Travel & Outdoors` +
`category_Video Games` +
`category_Web Series & TV Shows`+
category_Wellness +
`category_Writing & Publishing`,
data = .) %>%
summary()##
## Call:
## lm(formula = log_percent ~ category_Art + category_Audio + `category_Camera Gear` +
## category_Comics + category_Culture + `category_Dance & Theater` +
## category_Education + `category_Energy & Green Tech` + category_Environment +
## `category_Fashion & Wearables` + category_Film + `category_Food & Beverages` +
## `category_Health & Fitness` + category_Home + `category_Human Rights` +
## `category_Local Businesses` + category_Music + `category_Phones & Accessories` +
## category_Photography + `category_Podcasts, Blogs & Vlogs` +
## category_Productivity + `category_Tabletop Games` + category_Transportation +
## `category_Travel & Outdoors` + `category_Video Games` + `category_Web Series & TV Shows` +
## category_Wellness + `category_Writing & Publishing`, data = .)
##
## Residuals:
## Min 1Q Median 3Q Max
## -8.8126 -1.3999 0.2939 1.5251 15.4036
##
## Coefficients: (1 not defined because of singularities)
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 3.02390 0.05258 57.511 < 2e-16 ***
## category_Art -0.07755 0.07667 -1.011 0.311786
## category_Audio 1.18348 0.09849 12.016 < 2e-16 ***
## `category_Camera Gear` 0.85967 0.11444 7.512 5.96e-14 ***
## category_Comics 0.57464 0.07773 7.392 1.47e-13 ***
## category_Culture -0.19467 0.10333 -1.884 0.059583 .
## `category_Dance & Theater` 0.20605 0.07592 2.714 0.006652 **
## category_Education -1.42630 0.09337 -15.275 < 2e-16 ***
## `category_Energy & Green Tech` -0.74855 0.12350 -6.061 1.36e-09 ***
## category_Environment -1.22305 0.07798 -15.685 < 2e-16 ***
## `category_Fashion & Wearables` 0.21201 0.07642 2.774 0.005534 **
## category_Film -0.07647 0.06838 -1.118 0.263435
## `category_Food & Beverages` -0.46595 0.08340 -5.587 2.33e-08 ***
## `category_Health & Fitness` -0.45474 0.07657 -5.938 2.90e-09 ***
## category_Home 0.48864 0.07969 6.132 8.76e-10 ***
## `category_Human Rights` -0.47304 0.09552 -4.952 7.37e-07 ***
## `category_Local Businesses` -1.25713 0.07594 -16.554 < 2e-16 ***
## category_Music 0.26527 0.07217 3.675 0.000238 ***
## `category_Phones & Accessories` -0.58907 0.07800 -7.552 4.38e-14 ***
## category_Photography -0.07724 0.08403 -0.919 0.357966
## `category_Podcasts, Blogs & Vlogs` 0.15737 0.14039 1.121 0.262312
## category_Productivity -0.41073 0.08571 -4.792 1.66e-06 ***
## `category_Tabletop Games` 1.12962 0.10613 10.644 < 2e-16 ***
## category_Transportation -0.42912 0.09320 -4.604 4.15e-06 ***
## `category_Travel & Outdoors` 0.58253 0.07948 7.329 2.37e-13 ***
## `category_Video Games` -1.52121 0.07792 -19.522 < 2e-16 ***
## `category_Web Series & TV Shows` -0.44163 0.08342 -5.294 1.20e-07 ***
## category_Wellness -0.76199 0.10714 -7.112 1.16e-12 ***
## `category_Writing & Publishing` NA NA NA NA
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 2.347 on 37758 degrees of freedom
## Multiple R-squared: 0.073, Adjusted R-squared: 0.07234
## F-statistic: 110.1 on 27 and 37758 DF, p-value: < 2.2e-16We can see that almost all of our categories are significant values as the p-value is less than 0.05. More importantly, our top five categories as mentioned in the initial analysis are all significant as well. The next following regressions will focus on them to understand their success rates.
#top 5 multiple regression
indiegogo_dummy %>%
filter(funds_raised_100_percent > 0) %>%
lm(formula = log_percent ~
category_Audio +
`category_Tabletop Games` +
`category_Camera Gear` +
`category_Travel & Outdoors` +
category_Home,
data = .) %>%
summary()##
## Call:
## lm(formula = log_percent ~ category_Audio + `category_Tabletop Games` +
## `category_Camera Gear` + `category_Travel & Outdoors` + category_Home,
## data = .)
##
## Residuals:
## Min 1Q Median 3Q Max
## -8.8126 -1.4283 0.3625 1.6900 15.4036
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 2.67247 0.01330 200.97 <2e-16 ***
## category_Audio 1.53491 0.08640 17.77 <2e-16 ***
## `category_Tabletop Games` 1.48106 0.09543 15.52 <2e-16 ***
## `category_Camera Gear` 1.21111 0.10504 11.53 <2e-16 ***
## `category_Travel & Outdoors` 0.93397 0.06253 14.94 <2e-16 ***
## category_Home 0.84008 0.06280 13.38 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 2.406 on 37780 degrees of freedom
## Multiple R-squared: 0.02537, Adjusted R-squared: 0.02524
## F-statistic: 196.7 on 5 and 37780 DF, p-value: < 2.2e-16On this regression, we can see that the our top 5 categories have a positive value, meaning it is performing above the average and have the highest success rates.
Campaign duration
The next regression is log_percent ~ date_diff. Here we want to see if there is a relation between how successful your project is and if the campaign duration is relevant.
indiegogo %>%
filter(funds_raised_100_percent > 0) %>%
lm(formula = log_percent ~ date_diff,
data = .)%>%
summary()##
## Call:
## lm(formula = log_percent ~ date_diff, data = .)
##
## Residuals:
## Min 1Q Median 3Q Max
## -8.5779 -1.3052 0.3246 1.4791 31.5757
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 4.5777115 0.0262133 174.63 <2e-16 ***
## date_diff -0.0432129 0.0005766 -74.94 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 2.273 on 37784 degrees of freedom
## Multiple R-squared: 0.1294, Adjusted R-squared: 0.1294
## F-statistic: 5616 on 1 and 37784 DF, p-value: < 2.2e-16We can see that both values are significant and thus have a correlation. In this case, we see a negative correlation which means that the shorter amount of time you have your campaign duration, the more successful it will be.
indiegogo_dummy %>%
lm(formula = date_diff ~
category_Audio +
`category_Tabletop Games` +
`category_Travel & Outdoors` +
`category_Camera Gear` +
category_Home,
data = .)%>%
summary()##
## Call:
## lm(formula = date_diff ~ category_Audio + `category_Tabletop Games` +
## `category_Travel & Outdoors` + `category_Camera Gear` + category_Home,
## data = .)
##
## Residuals:
## Min 1Q Median 3Q Max
## -43.23 -12.23 0.77 17.77 677.77
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 42.2260 0.1071 394.400 < 2e-16 ***
## category_Audio -7.2746 0.6849 -10.622 < 2e-16 ***
## `category_Tabletop Games` -14.6056 0.7377 -19.800 < 2e-16 ***
## `category_Travel & Outdoors` -8.3247 0.5071 -16.417 < 2e-16 ***
## `category_Camera Gear` -5.4665 0.8344 -6.551 5.77e-11 ***
## category_Home -7.3119 0.4925 -14.848 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 19.89 on 40033 degrees of freedom
## Multiple R-squared: 0.0229, Adjusted R-squared: 0.02278
## F-statistic: 187.7 on 5 and 40033 DF, p-value: < 2.2e-16We create a regression with date_diff with our top five categories and see that all values are significant. Check out the figure below to show it with values filled in.
indiegogo %>%
filter(funds_raised_100_percent > 0,
date_diff <= 120,
category %in% c("Audio", "Tabletop Games","Travel & Outdoors", "Camera Gear", "Home")) %>%
ggplot(mapping = aes(log_percent, date_diff)) +
geom_point() +
geom_smooth(method = "lm") +
#ylim(0,120)+
#xlim(-5,10) +
geom_hline(yintercept = 60, color = "red") +
#geom_vline(xintercept = 45, color = "red") +
geom_hline(yintercept = 30, color = "red") +
geom_vline(xintercept = -3.5, color = "red") +
geom_vline(xintercept = 4.6, color = "red")## `geom_smooth()` using formula 'y ~ x'
# log_percent value when date_diff = 60
exp(-3.5)## [1] 0.03019738# log_percent value when date_diff = 30
exp(4.6)## [1] 99.48432The model shows is that longer your campaign goes for, the less successful it will be. The output we see if date_diff = 60, is a 3% success rate, while, if date_diff = 30, the success rate is 99.5%. Obviously, this is the model, and it is not the most accurate, but the general idea is to keep your campaign in the 30-day range to stay relevant.
Goal Achieved
The next model is goal_achieved and the top 5 categories. The goal_achived variable determines whether or not a project reached their goal.
indiegogo_dummy %>%
lm(formula = goal_achieved ~
category_Audio +
`category_Tabletop Games` +
`category_Travel & Outdoors` +
`category_Camera Gear` +
category_Home,
data = .)%>%
summary()##
## Call:
## lm(formula = goal_achieved ~ category_Audio + `category_Tabletop Games` +
## `category_Travel & Outdoors` + `category_Camera Gear` + category_Home,
## data = .)
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.5214 -0.2050 -0.2050 -0.2050 0.7950
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 0.205016 0.002246 91.26 <2e-16 ***
## category_Audio 0.316371 0.014370 22.02 <2e-16 ***
## `category_Tabletop Games` 0.288927 0.015478 18.67 <2e-16 ***
## `category_Travel & Outdoors` 0.227634 0.010640 21.39 <2e-16 ***
## `category_Camera Gear` 0.201558 0.017507 11.51 <2e-16 ***
## category_Home 0.196035 0.010333 18.97 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.4174 on 40033 degrees of freedom
## Multiple R-squared: 0.03842, Adjusted R-squared: 0.0383
## F-statistic: 319.9 on 5 and 40033 DF, p-value: < 2.2e-16Here as well, we can see a significant relation amongst all of our categories. A cool finding was that the value is actually the mean of the success_rate, which we found on Figure 6.
Log_amount
The final regression is the log_amount and goal_achived.
indiegogo %>%
filter(funds_dollars > 0) %>%
lm(formula = log_amount ~ goal_achieved,
data = .)%>%
summary()##
## Call:
## lm(formula = log_amount ~ goal_achieved, data = .)
##
## Residuals:
## Min 1Q Median 3Q Max
## -6.665 -1.230 0.046 1.282 6.846
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 6.39498 0.01083 590.4 <2e-16 ***
## goal_achieved 3.44725 0.02164 159.3 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.827 on 37979 degrees of freedom
## Multiple R-squared: 0.4006, Adjusted R-squared: 0.4006
## F-statistic: 2.538e+04 on 1 and 37979 DF, p-value: < 2.2e-16We continue to see the values as signficant when compared to log_amount. But the main question is - what is the exact amount needed on average to be a successful project in these categories?
indiegogo_dummy %>%
filter(funds_dollars > 0) %>%
filter(goal_achieved == 1) %>%
lm(formula = log_amount ~
category_Audio +
`category_Tabletop Games` +
`category_Travel & Outdoors` +
`category_Camera Gear` +
category_Home,
data = .) %>%
summary()##
## Call:
## lm(formula = log_amount ~ category_Audio + `category_Tabletop Games` +
## `category_Travel & Outdoors` + `category_Camera Gear` + category_Home,
## data = .)
##
## Residuals:
## Min 1Q Median 3Q Max
## -4.9886 -1.4156 -0.1237 1.2802 7.2013
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) 9.48733 0.02259 420.004 <2e-16 ***
## category_Audio 1.75509 0.09230 19.015 <2e-16 ***
## `category_Tabletop Games` -0.06056 0.10175 -0.595 0.552
## `category_Travel & Outdoors` 1.50288 0.07545 19.919 <2e-16 ***
## `category_Camera Gear` 1.62577 0.12602 12.901 <2e-16 ***
## category_Home 1.71525 0.07595 22.585 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 1.901 on 9510 degrees of freedom
## Multiple R-squared: 0.1137, Adjusted R-squared: 0.1133
## F-statistic: 244.1 on 5 and 9510 DF, p-value: < 2.2e-16The code below shows us the average amount needed to reach success in each category (in $).
audio_amount = round(exp(9.97353+1.26918), digits = 2)
print(paste("Audio: ", audio_amount), quote = FALSE)## [1] Audio: 76321.5tabletop_amount = round(exp(9.97353-0.55734), digits = 2)
print(paste("Tabletop Games: ", tabletop_amount), quote = FALSE)## [1] Tabletop Games: 12285.68travel_amount = round(exp(9.97353+1.40227), digits = 2)
print(paste("Travel & Outdoors: ", travel_amount), quote = FALSE)## [1] Travel & Outdoors: 87186.08camera_amount = round(exp(9.97353+1.18342), digits = 2)
print(paste("Camera Gear: ", camera_amount), quote = FALSE)## [1] Camera Gear: 70048.98home_amount = round(exp(9.97353+1.59241), digits = 2)
print(paste("Home: ", home_amount), quote = FALSE)## [1] Home: 105444.5